Page 229 - MEGIN Book Of Abstracts - 2023

Parkinson's disease. Gradient-boosted tree learning Laboratory, Children's Hospital, Helsinki University Hospital
was applied in combination with feature importance and Helsinki University, Helsinki, Finland; Biomag Laboratory,
analysis to generate and understand out-of-sample HUS Medical Imaging Center, Helsinki University Hospital,
predictions. Helsinki University, and Aalto University School of Science,
Helsinki; Department of Neurology, Helsinki University Hos-
RESULTS A few features sufficed for making accurate pital and Department of Clinical Neurosciences (Neurology),
predictions. A model operating on five coherence University of Helsinki, Helsinki, Finland
features, for example, achieved correlations of r > 0.8
between actual and predicted outcomes. Coherence ABSTRACT Despite optimal oral drug treatment, about
comprised more information in less features than sub- 90% of patients with Parkinson's disease develop mo-
thalamic power, although in general their information tor fluctuation and dyskinesia within 5-10 years from
content was comparable. Both signals predicted aki- the diagnosis. Moreover, the patients show non-motor
nesia/rigidity reduction best. The most important local symptoms in different sensory domains. Bilateral deep
feature was subthalamic high-beta power (20-35 Hz). brain stimulation (DBS) applied to the subthalamic
The most important connectivity features were sub- nucleus is considered the most effective treatment in
thalamo-parietal coherence in the very high frequency advanced Parkinson's disease, and it has been sug-
band (>200 Hz) and subthalamo-parietal coherence gested to affect sensorimotor modulation and relate to
in low-gamma band (36-60 Hz). Successful prediction motor improvement in patients. However, observations
was not due to the model inferring distance to target or on the relationship between sensorimotor activity and
symptom severity from neuronal oscillations. clinical improvement have remained sparse. Here, we
studied the somatosensory evoked magnetic fields in
CONCLUSION This study demonstrates for the first 13 right-handed patients with advanced Parkinson's
time that neuronal oscillations are predictive of DBS disease before and 7 months after stimulator implanta-
outcome. Coherence between subthalamic and parietal tion. Somatosensory processing was addressed with
oscillations are particularly informative. These results magnetoencephalography during alternated median
highlight the clinical relevance of inter-areal synchrony nerve stimulation at both wrists. The strengths and the
in basal ganglia-cortex loops and might facilitate fur- latencies of the ~60-ms responses at the contralateral
ther improvements of DBS in the future. primary somatosensory cortices were highly variable
but detectable and reliably localized in all patients. The
Keywords: Deep brain stimulation, Machine learning, response strengths did not differ between preoperative
Neuronal oscillations, Parkinson's disease, Subthalamic and postoperative DBSON measurements. The change
nucleus in the response strength between preoperative and
postoperative condition in the dominant left hemi-
Brain stimulation (2022), Vol. 15, No. 3 (35568311) (2 sphere of our right-handed patients correlated with the
citations) alleviation of their motor symptoms (p = .04). However,
the result did not survive correction for multiple com-
parisons. Magnetoencephalography appears an effec-
Modulation of sensory cortical activity by deep tive tool to explore non-motor effects in patients with
brain stimulation in advanced Parkinson's disease Parkinson's disease, and it may help in understanding
(2022) the neurophysiological basis of DBS. However, the
high interindividual variability in the somatosensory
Korsun, Olesia; Renvall, Hanna; Nurminen, Jussi; Mäkelä, responses and poor tolerability of DBSOFF condition
Jyrki P; Pekkonen, Eero warrants larger patient groups and measurements also
in non-medicated patients.
Department of Neuroscience and Biomedical Engineering,
School of Science, Aalto University, Espoo; Motion Analysis

 ontents  Index 208
C